varunananth2003 / java-neural-network

A fully functioning neural network made from scratch in Java without the use of any external libraries. Comes with extended functionality such as L2 regularization, Mini-batch gradient descent, variable activation functions, etc. Fully commented for ease of understanding! Now imported to IntelliJ.

Java Neural Network

These files contain everything that is needed to create, train, use, save, read, and edit a simple Neural Network!

The code is fully commented, so you can see how a Neural network works from the ground up. The class "Main.java" contains example creation, training, saving, reading, and testing of a network.

Current Features: Unlimited layer expansion (theoretically), Backpropagation, Variable activation functions, Variable cost function, Stochastic, Batch, or Mini-batch gradient descent, L2 regularization, Trained network save/read

Template NN Save file:

[COST FUNCTION]

[# OF LAYERS]

[NEURONS IN CURRENT LAYER (n)] [NEURONS IN NEXT LAYER (m)] [ACTIVATION FUNCTION]

[WEIGHT ARRAY FOR NEURON 1 (length m)]

[BIAS FOR NEURON 1]

...

[WEIGHT ARRAY FOR NEURON n (length m)]

[BIAS FOR NEURON n]

[NEURONS IN CURRENT LAYER (m)] [NEURONS IN NEXT LAYER (k)] [ACTIVATION FUNCTION]

...

Meta info: This project was completed without the use of external libraries and is "pure java" This whole project took about one week to make, another to refine, and another to comment. Let me know if you want to see any features or functionality added. Enjoy!